View / X lost its CEO, but it hasn’t lost Musk

It’s been a bad week for Elon Musk’s xAI. Despite its Grok chatbot becoming the world’s most advanced model, according to a series of benchmarks, the company was hit with multiple public relations setbacks. First, its Grok chatbot made out-of-the-blue antisemitic comments on X. Then, X CEO Linda Yaccarino resigned. Later, Grok users noticed the chatbot was consulting Musk’s personal tweets before weighing in on hot button political issues.

It’s emblematic of Musk’s career lately. Astonishing accomplishments in everything from space travel to neuroscience to autonomous driving to AI, all muted by his polarizing public persona.

There’s no question that Grok’s antisemitic posts were a mistake. Musk admitted as much and said the company is trying to rectify the issue. The chatbot’s search for Musk’s personal views on the topic likely falls into the same category. Grok made no secret that it was using Musk as a source, spelling it out in plain English in its list of reasoning steps.

The backlash was fierce. The Atlantic, for instance, argued Musk and xAI researchers had “communicated their values [to Grok] and given it clear instructions. That the machine has read them and responded by turning into a neo-Nazi speaks volumes.”

In reality, it’s possible to communicate one set of values to an LLM and get opposing outputs. Just ask Google. It wanted its Gemini model to better represent the diversity of the real world so that if a person prompted a photo of a doctor, they wouldn’t get a white male 100% of the time. Instead, the model ended up depicting World War II Nazis as Black.

When xAI pledged to create a model that is not “politically correct,” it was setting itself up for some embarrassment. Humans — like Musk himself — often have trouble walking the tight rope between being politically incorrect and offensive. And Grok, after consulting Musk’s tweets, said it supported Israel in the Gaza conflict. (Following the logic of his critics, that makes him one of the few antisemitic supporters of Israel).

The challenge of making chatbots “think” a certain way is, at this point, alchemy. That’s true for xAI, Google, Anthropic, and others, and will remain an issue until someone solves interpretability, the effort to scientifically understand why models behave the way they do. That’s what Missouri Attorney General Andrew Bailey, who’s suing Google, Microsoft, OpenAI, and Meta for bias against President Donald Trump, will likely find in his investigation.

Where does that leave Musk? Last September, which is ancient history when it comes to tech these days, I wrote about how he was positioning himself to take the lead in AI, and that X could become a distribution platform for Grok. Quite a few people told me I had gone nuts. xAI eventually acquired X, and yesterday, Grok became the leading AI model in the world. This is purely about speed. Musk just moves faster than anyone else, and he will not take his foot off the gas now. To those of you who owe me a drink, you know who you are.